I’ve watched a lot of really interesting philosophical YouTube videos about the dangers of AI lately. It’s an intriguing subject. Maybe you have already heard a lot of about the possible danger of AI reaching a level in the future where it may be smarter than humans. Imagine if it can also repair and even replicate itself. If we are not careful about the timing of this, it could get out of hand and completely wipe out humanity.
There’s this excellent analogy with ants. Ants can’t possibly understand anything we do and can’t protect themselves against us, but normally we wouldn’t want to do ants any harm either.
Until we need to build a house where an anthill is sprawling.
One of the YouTube videos I watched was Answers With Joe’s video about The Dark Side Of The Singularity. Joe makes a point that he’s actually more worried about the economic singularity than the climate changes. An economic singularity means that AI makes for so much automation that a large portion of humans on the planet will be unemployed. It could get so bad that we may have to rethink how we manage the economy in our societies, create a minimum income, or maybe even change how money works.
It’s a great video and I recommend you watch it:
I’ve had this vague idea that germinated in my mind a few months ago, and Joe’s video made me ponder it a lot more. The thing is, there’s something I’m even more worried about than Joe’s economic singularity, and that’s the creative singularity. It scares the crap out of me.
What do I mean by a creative singularity?
As the AI gets better at thinking and learning, I predict that it will at one point surpass everything we can do on a creative level. It may not come at once as being creative can mean a lot of things. There’s painting, composing music, designing cars, making the next bridge more efficient, or even crocheting. Some of these things may be harder to learn for an AI than others, but I have no doubt that eventually it will figure it out.
The problem is, it will not just be able to do it as we do. I’m not talking about it sitting down next to me to crochet the trousers for the blouse I’m working on right now. Thanks, mate. What I’m talking about is how it would completely crush us. It would learn to do things so much better, it would make your smile slowly disappear. And then it gets even worse. Much worse.
It would be able to compose songs much better than anything Bach, Mozart or Beethoven have even done, including the perfect instruments for the various parts. Maybe it would even invent new instruments you have never heard before, but which sound awesome where used. It will be able to compose in any genre at all. Sing like Elvis Presley too? Sure, it can emulate him and you won’t be able to tell the difference.
At one point I watched another YouTube video about the dangers of AI, and here the narrator showed an example of an AI playing the piano where the AI even figured out how to mimic the piano sound itself.
It’s a part that starts at 4:14 in this video by LEMMiNO:
Mind-boggling, isn’t it? And that’s just the beginning.
I also have absolutely no doubt that some time in the future, AI will make your movies and television shows too, and they will also be so much better. Better direction, the stories will be fantastic, and the music will be astonishing. It can emulate any old actor you remember and just as with above example with Elvis Presley, you won’t be able to tell the difference anymore.
That might actually be both a curse and a blessing. It’s bad for human filmmakers that can’t even begin to compete, but great for viewers who know they can rely on the next season of Star Trek will have awesome episodes that all look incredibly expensive. No more Star Trek: Discovery.
I imagine that a separate wave of creativity will be born and worshiped where created by humans will be a respected label that people will actually care for. Kind of like how vinyl is gaining popularity again these days. Some may even be fine with AI being able to create much better art than us, but personally I don’t think I could ever settle with that.
As a SID composer on the Commodore 64, I’ve always been envious of better composers, trying to reach the same level and maybe in the process manage to create something they never thought of. Creating a nice and unique composition is fun by itself, but the competition and the feeling of being on the edge is part of what gives me creative satisfaction.
What is the point in spending days creating something that the AI in my house can create much better on its own in just a few minutes? Why would you need me anymore?
But. I am not a specialist and have not educated myself on the subject as you have, but I’ll approach by logic.
How do AI’s get trained to know what is good and what is bad entertainment? By looking/listening to/analyzing/sampling works that humans have already created and humans have rated good/bad/entertaining/not entertaining etc.
So I would assume that an AI would be able to mimic stuff that humans like, but at best it would be a caricature or mashup of what it is known that we humans like, based on ratings.
So to create a super-masterpiece in a certain medium, let’s say a film, the AI only would only know about situations and plots and stories that we humans have already invented. If it did mashups of randomly selected situations, let’s say in the romantic film category, a million of those would make no sense to us or have strange plot holes watching the resulting film. If the AI begins creating new and never-before-seen situations in a film, it has no idea if this new hitherto-unknown material would be rated good and entertaining by humans, unless humans actually sat down and watched it.
On the subject of music, if the AI has no rating of what scales and harmonies and rhythms we like, any random garble would be just as valuable to it as any other random garble. If it knows that this and this classical symphony are the ‘absolutely best’ ever created according to humans, it ‘absolutely best’ the AI can create, according to itself, would be a caricature or a mashup of those same ‘absolute best’ musical works. If it introduced mutations and created something different or unique not present in those known masterpieces, how would it know that this new composition would be better or evoke more/stronger emotions in us humans? Unless humans had time to sit down and listen to every new resulting AI composition and then judge it. And then progress is still limited by human time and human taste. And we humans generally like new stuff that reminds us of previous stuff that is already familiar to us. So AI composed music that we would deem enjoyable, would be mashups or caricatures of already known material.
The first thing I think of here is how humans are already terribly good at emulating and approximating previously great stuff, especially when we’re doing it deliberately (see Star Wars: The Force Awakens). I also think of the YouTube video I watched a few years ago about everything being a copy of a copy.
Other than that, you’re probably right that the AI would at first mimic what humans made, constantly comparing their own creations with their gems to determine if they’re on the right track. But actually I only regard that as the first version. In the second version, which I imagine the AI may very well develop on its own, this is slowly replaced with a subjectivity algorithm. Maybe it will retain a big smattering of the algorithm from the first version, but as time goes on, that shrinks and gets overshadowed by the second version. The AI learns to trust its own creative gut instincts better. It no longer needs to compare its creations to existing human stuff because it now just sort of knows what it likes.
That’s when we’re going to see some really crazy stuff – brilliant but also quite original.